The second speaker in this IAMCR 2024 session is Juan Antonio Guevara, whose interest is in polarisation in the 2023 Spanish general elections. His focus here is especially in affective polarisation, which can mean different things depending on how the idea is conceptualised. Here, polarisation is approached through a ‘fuzzy-set’ approach drawn from mathematics.
This recognises that reality is not black and white, but that individuals may have different levels of affiliation towards a variety of political parties or positions; it measures the individual’s level of affiliation towards both poles of several possible scales of affiliation. These can then be combined into a single measure of polarisation on a single scale. This also addresses the issue that polarisation might still be a problem when the vast majority of a population holds the same values, but when these values actually represent an extreme perspective.
For larger datasets drawn from digital and social media, the task of assessing individuals’ positioning across these various scales can be approached by employing transformer models that generate vector representations of their content. This might work on the post texts, the sentiments contained in them, or other possible features.
Applying this to the 2023 elections in Spain, a snap election called by the Spanish Prime Minister. The dataset here was drawn from CrowdTangle, and covers public pages, public groups, and public verified profiles in Facebook that engaged in discussing the elections. The project applied the BERTopic model to the posts, performed a clustering analysis to determine key topics, and used RoBERTuito to identify sentiment in these posts.
Key topics identified by this process addressed candidates, activism, opposition, general debate, voting, and electoral participation; posts were predominantly neutral, but some sentiment and a small level of hateful and aggressive speech was also present. Affective polarisation was relatively low, but this is also a function of the specific Facebook spaces covered by the dataset. Posts in the activism topic showed the greatest level of polarisation.
